The Gulf Long-term Follow-up (GuLF) Study is a prospective cohort study designed to examine human health effects following the 2010 Deepwater Horizon oil spill in the Gulf of Mexico. The GuLF Study includes 32,608 adults ≥ 21 years of age who participated in oil spill response and cleanup work (N = 24,937) and those who trained for potential work but were not hired (N = 7671) [18]. Participants from across the U.S., but largely from the Gulf region, were enrolled between March 2011 and March 2013 and completed a 30- to 60-minute computer-assisted telephone enrollment interview. The interview collected information on oil spill response and cleanup activities, and demographic, socioeconomic, occupational, lifestyle, and health information [18]. Study questionnaires can be found at www.niehs.nih.gov/gulfstudy. Participants from eastern Texas, Louisiana, Mississippi, Alabama, and Florida (Gulf Coast region) were invited to have a home visit (N = 26,828). A total of 11,119 completed the home visit (May 2011-May 2013) which included collection of biological samples, functional measures (e.g., blood pressure, lung function), and anthropometrics, and examiner-administered questionnaire data, including mental health screeners. Home visit participants provided written informed consent. The study was approved by the Institutional Review Board of the National Institutes of Health.

For this analysis, we were interested in two primary mental health symptoms: depression and psychological distress. We used the original questionnaires of Patient Health Questionnaire-9 (PHQ-9) for depression and Kessler Quick Inventory of Distress (K6) for psychological distress as they have strong validity and reliability (questionnaire is available at https://epishare.niehs.nih.gov/studies/GuLF/).

Specifically, the PHQ-9 was used to estimate current depression at the time of home visit by assessing symptoms that occurred over the past two weeks. PHQ-9 is the 9-item depression severity measure adapted from the full PHQ. Each item of PHQ-9 is scored from 0 (not at all) to 3 (nearly every day) and the total score can range from zero to 27. Studies showed that compared to an independent structured mental health professional interview, a PHQ-9 score ≥10 yielded a sensitivity of 88% and a specificity of 88% for major depression [19]. We used this threshold to categorize the participants as “depressed” and “not depressed”. There were 10,217 participants who completed PHQ-9 and had complete data for all other covariates.

We used the K6 to evaluate current psychological distress at the time of home visit [20]. K6 has been widely used and validated to assess non-specific psychological distress symptoms such as feelings of nervousness, hopelessness, and worthlessness. Except for “don’t know” and “refused”, the responses range from “none of the time” to “all of the time”. The six items in K6 are summed to a score ranging from zero to 24. Previous studies have demonstrated that K6 has excellent internal consistency and reliability. Prior validation studies of the K6 against clinical structured diagnosis reported an accuracy of 0.92 at a cut‐point ≥ 13 [20, 21]. Again, we dichotomized the scores of psychological distress based on this threshold. The K6 was added to the home visit sometime after the home visits had started. Thus, there were 8,765 participants who completed the K6 and had complete information on all relevant covariates.

We obtained solar radiation exposure estimates (Watt/m2) from the newly updated Daily Surface Weather Data (Daymet Version 4 R1) database [22]. Daymet is a high-quality research product of the Environmental Sciences Division at Oak Ridge National Laboratory and is supported by the National Aeronautics and Space Administration (NASA) through the Earth Science Data and Information System (ESDIS) and the Terrestrial Ecology Program that combines measured data with models to achieve 1 km x 1 km grid estimates of climate factors [22]. The estimates from Daymet can have a correlation with observations up to a high R2 of 98.9% [23]. Daymet provides information on daily minimum temperature, maximum temperature, precipitation, shortwave radiation, vapor pressure, snow water equivalent, and day length from 1980 to 2021. The gridded datasets were spatially linked to participants’ geocoded home visit residential addresses. We calculated the average solar radiation (SRAD) exposure over the seven (SRAD7), 14 (SRAD14), and 30 (SRAD30) calendar days before each participant’s home visit. SRAD7, SRAD14, and SRAD30 were categorized into quartiles (Q1, Q2, Q3, and Q4) based on the empirical distribution among the analytic sample. Temperature and humidity were also obtained from the Daymet database.

We analyzed associations between mental health status and residential solar radiation exposures in the past seven (SRAD7), 14 (SRAD14), and 30 days (SRAD30) prior to the home visit. We generated maps showing the average values of the SRAD7, SRAD14, and SRAD30 that participants in each county/Parish were exposed to. We used crude and adjusted generalized linear mixed models to estimate prevalence ratios (PR) and 95% confidence intervals (CI) for associations between solar radiation and depression and distress. We developed three models: 1) crude models; 2) models adjusted for age groups (20-40, 40-60, and >60), self-reported race (White, Black, and other), ethnicity (Hispanic, non-Hispanic), self-reported sex (female, male), education (less than high school, high school, some college, and college or greater), employment status at the time of enrollment (yes, no), ever alcohol consumption (yes, no), ever cigarette smoking (yes, no), Deepwater Horizon oil spill cleanup worker status (yes, no), season (spring, summer, fall, and winter), residence in a county or Parish abutting the Gulf of Mexico (“proximity”; yes, no), and state of residence (as a random effect); and 3) models additionally adjusting for temperature and relative humidity with corresponding time lags [10, 24]. We did not adjust for income because of a high proportion of missing values (N = 787, 7.1%). Instead, we carried out a sensitivity analysis and found similar results in models with and without adjustment for income where all other covariates were controlled (Supplementary Table A1).

We used a directed acyclic graph to show the presumed relationships among potential confounders (Supplementary Fig. A1). In addition, we conducted stratified analyses to evaluate effect modification by season, age group, and sex [25]. For the age we stratified the participants at age 50, the approximate median age of the cohort, which gave us reasonable sample size in each age stratum. We also conducted a sensitivity analysis among participants who had complete data for both outcomes and all other covariates (Supplementary Fig. A2). Statistical significance was defined as p value < 0.05. All analyses were conducted in R version 4.2.1.